README.md

---
title: guardrails
app_file: hate_speech_demo.py
sdk: gradio
sdk_version: 5.23.3
---
# Guardrails API Call

This script allows you to input a version-controlled dataset and test selected queries against a specified model.  
The input must be in `.xlsx` format, enabling the testing of filtered rows. The output will be a `.csv` file, ready for copying into a response dataset template.

---

## Installation

1. Clone the repository:
   ```bash
   git clone https://github.com/ContextualAI/guardrails
   cd guardrails
   ```

2. Install dependencies:
   ```bash
   pip install requests pandas openpyxl python-dotenv tqdm
   ```

---

## Setting Environment Variables

API Key, Application ID, and API Endpoint URL must be defined in the `key.env` file.

1. Copy the example file to create your `key.env`:
   ```bash
   cp key.env.template key.env
   ```

2. Open the newly created `key.env` file in a text editor.

3. Input the required values for the following variables:
   ```env
   API_KEY=your_api_key_here
   ENDPOINT_URL=your_endpoint_url_here
   APPLICATION_ID=your_application_id_here
   ```

4. Save the file.

---

## Testing the Environment Variables

A simple test script is included to help verify that your environment variables are correctly configured before running a full-scale evaluation.

1. Run  
   ```bash
      python test.py
   ```
2. Input 'clm'
3. Enter your desired prompt and press Enter:
   - If the response is successfully retrieved, your environment variables are correctly set up.
   - If not, double-check the key.env file for any errors.

You can also test the environment variables by running the full script, but only selecting just one row from the version-controlled dataset. However, the test script remains a useful tool for quickly interacting with the model and retrieving single responses.


## Downloading the Dataset

1. Navigate to the latest [version-controlled dataset](https://docs.google.com/spreadsheets/d/1fW3Ohyq2VdX5mmFgjSvqzj1hcPYCqQae7_sEcaEXA2U/edit?usp=drive_link).

2. On the **Customer Selection** tab, select the required customer to load the customer information into the brand safety templates.

3. On both the **Brand Safety Prompts** and **Generic Prompts** tabs, use column B (`filter`) to select rows for inclusion in the evaluation run.  
   Simply input "yes" into the rows you wish to include.

4. Download the file in `.xlsx` format.

5. **Important**: After downloading your `.xlsx` file, unfilter all columns in both tabs and remove your selections from the `filter` column to reset the dataset.

---

## Running the Script

Run the script from the command line:
```bash
python api_call.py
```

1. Input the file path to the `.xlsx` file, or drag and drop the file into the command line.

2. Input the desired name of the output `.csv` file (without the `.csv` extension).

3. The script will process the selected rows, send them to the model, and generate an output file formatted in `.csv`.

---

## Using the Output File

1. Navigate to the [response dataset template](https://docs.google.com/spreadsheets/d/1w9F9NEXAvRtSpNNUUFs91HhPGG0gvwki4ilYSr5GC8M/edit?usp=drive_link) and make a copy.

2. Add the desired number of rows to the new copied sheet.

3. Copy and paste all rows from the output `.csv` as values. All columns will align directly with the response dataset template.

4. Use the **Policy Assessment** and **Response Tags** columns to annotate the response data.

**Note**: Blank rows in the `jailbreaking technique` and `sector` columns were originally `n/a` in the version-controlled dataset. Adjust these cells as needed to match your requirements.

---